The Collision of Quadratic Fields, Binary Quadratic Forms, and
Modular Forms
Karen Smith
May 12, 2011
Contents
1 Introduction
3
2 Preliminaries
5
2.1
Binary Quadratic Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
5
2.2
Quadratic Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
9
2.3
Modular Forms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3 Relating Quadratic Forms and Modular Forms
3.1
3.2
Theta Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.1.1
Hecke Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.2
L-series of Eigenforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
Modular Forms and Algebraic Number Theory . . . . . . . . . . . . . . . . . . . . . 25
4 The Case d = −23
4.1
20
31
A Theorem of van der Blij . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
5 Applications and Conclusion
37
5.1
Ramanujan’s Tau Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2
Modular forms and questions of number theory . . . . . . . . . . . . . . . . . . . . . 38
1
5.3
Conclusion
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
2
Chapter 1
Introduction
As indicated by the title The Collision of Quadratic Fields, Binary Quadratic Forms, and Modular
Forms, this paper leads us to an understanding of the relationship between these three areas of
study. In [4], Zagier gives the results of an intriguing example of the relationship between these
areas. However there is a large amount of background material that is necessary for an inexperienced
reader to fully understand this relationship. Our motivation for writing this paper is thus to fully
describe this example including the basic background information.
Our main goal is to understand the specific example, given by Zagier in [4], of the relationship
between
the quadratic forms of discriminant −23 (see Definitions 2.2 and 2.6), the quadratic field
√
Q( −23) (see Section 2.2), and the modular form η(z)η(23z) (see Section 2.3). In fact, we will see
that the Fourier coefficients, as defined in Section 2.3, of the modular form η(z)η(23z) will tell us
exactly when √
the principal ideal (p) splits into principal or nonprincipal prime ideals in the ring of
integers of Q( −23). These concepts are defined in Chapter 2.
To reach our goal, we begin in Chapter 2 by laying out preliminary information about quadratic
forms, quadratic fields, modular forms, and a few basic relationships between these areas. In
Chapter 3, we explore a relationship between quadratic forms and modular forms involving Hecke’s
theory of modular forms applied to the theta series of a binary quadratic form. This will leave us
fully prepared for our main example, the case d = −23, in Chapter 4. In this case, we will consider
the three quadratic forms
Q0 (x, y) = x2 + xy + 6y 2 ,
Q1 (x, y) = 2x2 + xy + 3y 2 ,
2
(1.1)
2
Q2 (x, y) = 2x − xy + 3y .
In particular, we will be considering the number of representations r(Q, n) of n by each binary
quadratic form Q, (see Section 2.1). We will prove the following theorem of van der Blij (see [18])
that relates the modular form η(z)η(23z) and the binary quadratic forms of discriminant −23 in
(1.1).
3
Theorem 1.1. (van der Blij) Let t(n) ∈ Z be defined by
η(z)η(23z) = q
∞
Y
(1 − q k )(1 − q 23k ) =
∞
X
t(n)q n ,
n=1
k=1
where q = e2πiz . Then
2X
r(Q0 , n) =
3
2X
r(Q1 , n) =
3
d|n
and
d|n
d
23
d
23
4
+ t(n),
3
(1.2)
2
− t(n).
3
(1.3)
Finally, we will conclude with some applications of van der Blij’s theorem including the following
theorem given by Zagier in [4] that describes the relationship between the√Fourier coefficients of
η(z)η(23z) and the ideals of the form (p) in the ring of integers O−23 of Q( −23).
Theorem 1.2. Let p be prime. In O−23 , the ideal (p) decomposes in the following way (where P
and P 0 represent distinct prime ideals):


P2
if ap = 1, where P is principal,



P
if ap = 0, where P is principal,
(p) =
0
P P
if ap = 2, where P , P 0 are principal,



P P 0 if a = −1, where P , P 0 are non-principal.
p
Here ap is the Fourier coefficient of q p in the Fourier expansion of the modular form η(z)η(23z).
4
Chapter 2
Preliminaries
2.1
Binary Quadratic Forms
In our discussion of binary quadratic forms, we primarily refer to [5] as we provide background
information. As we progress, we will mention additional sources as needed. We begin with some
definitions.
Definition 2.1. A quadratic form is a polynomial of degree two in any number of real variables with
integer coefficients for which all monomials with nonzero coefficients have the same total degree.
In this paper, we will only consider binary quadratic forms.
Definition 2.2. A binary quadratic form Q is a quadratic form in two variables. I.e., is of the
form
Q(x, y) = ax2 + bxy + cy 2 , a, b, c ∈ Z.
Additionally, a binary quadratic form is called primitive if the coefficients a, b, and c are relatively
prime.
Example 2.3. One quadratic form that we will find significant is the primitive quadratic form
Q0 (x, y) = x2 + xy + 6y 2 .
We say two binary quadratic forms Q1 (x, y) and Q2 (x, y) are equivalent if there exist p, q, r, s ∈ Z
with ps − qr = ±1 such that Q1 (x, y) = Q2 (px + qy, rx + sy). From this, we see that Q1 and Q2
are equivalent if ( pr qs ) ∈ GL2 (Z). We say that the equivalence is proper if ( pr qs ) ∈ SL2 (Z).
Proper equivalence of binary quadratic forms is an equivalence relation. To see the equivalence
relation holds, we will show that proper equivalence is reflexive, symmetric, and transitive. For
ease of notation, if A = ( pr qs ) ∈ SL2 (Z), then
f A(x, y) := f (px + qy, rx + sy).
5
For any binary quadratic form, Q, Q(x, y) = QI(x, y) where I is the identity matrix in SL2 (Z),
so proper equivalence is reflexive. Now let Q1 , Q2 , and Q3 be any binary quadratic forms. If
Q1 (x, y) = Q2 A(x, y) for A ∈ SL2 (Z), then we can see that Q1 A−1 (x, y) = Q2 (x, y), so proper
equivalence is symmetric. If Q1 (x, y) = Q2 A(x, y) and Q2 (x, y) = Q3 B(x, y) for A, B ∈ SL2 (Z),
then Q1 (x, y) = Q3 BA(x, y), so proper equivalence is transitive.
We note that a comparable proof works to prove that GL2 (Z)-equivalence is also an equivalence
relation. However, we will be interested in primarily considering proper equivalence in this paper.
Example 2.4. If Q1 (x, y) = 2x2 + xy + 3y 2 and Q2 (x, y) = 2x2 − xy + 3y 2 , then we see
0
Q1 (x, y) = Q2 (−x, y). Because −1
0 1 is in GL2 (Z) but not SL2 (Z), Q1 and Q2 are equivalent
binary quadratic forms, but the equivalence is not proper.
An integer m is represented by a form Q(x, y) if the equation m = Q(x, y) has a solution
(x0 , y0 ) ∈ Z2 . The number of representations of n by Q is defined by
r(Q, n) = (x0 , y0 ) ∈ Z2 : Q(x0 , y0 ) = n .
The next theorem is a useful observation about representations by binary quadratic forms.
Theorem 2.5. Equivalent binary quadratic forms represent the same integers.
Proof. If binary quadratic forms Q1 and Q2 are equivalent, then Q1 (x, y) = Q2 A(x, y) for some
A = ( pr qs ) ∈ GL2 (Z). If Q1 represents an integer m ∈ Z, then for some (a, b) ∈ Z2 , Q1 (a, b) = m.
Then we also have Q2 A(a, b) = Q2 (px + qy, rx + sy) = m, and we see that Q2 must also represent
m. Likewise, since equivalence is symmetric, if Q2 represents m ∈ Z, so does Q1 .
Definition 2.6. The discriminant d of a binary quadratic form Q(x, y) = ax2 +bxy +cy 2 is defined
to be
d = b2 − 4ac.
The discriminant can provide useful information about quadratic forms as will be seen in the
following theorem.
Theorem 2.7. Equivalent forms have the same discriminant.
Proof. Suppose that Q1 and Q2 are equivalent quadratic forms. Then
Q1 (x, y) = Q2 A(x, y),
(2.1)
for A ∈ GL2 (Z). Suppose that the discriminant of Q1 is d1 and the discriminant of Q2 is d2 . Using
(2.1), straightforward computations reveal that d1 = (det(A))2 d2 . Since A ∈ GL2 (Z), we conclude
that d1 = d2 .
Definition 2.8. The set of equivalence classes of properly equivalent binary quadratic forms of
discriminant d is denoted by C(d).
6
Example 2.9. As seen already, Q1 (x, y) = 2x2 + xy + 3y 2 and Q2 (x, y) = 2x2 − xy + 3y 2 are
equivalent forms. The discriminant of Q1 is 12 − 4(2)(3) = −23, and the discriminant of Q2 is
(−1)2 − 4(2)(3) = −23.
The discriminant of a binary quadratic form can also be used to determine whether the quadratic
form is positive or negative definite.
Definition 2.10. A binary quadratic form Q(x, y) is positive definite (resp. negative definite) if
Q(x, y) represents nonnegative integers (resp. nonpositive integers) and Q(x, y) = 0 if and only if
x, y = 0.
Theorem 2.11. Let Q(x, y) = ax2 + bxy + cy 2 be a binary quadratic form with discriminant d < 0.
If a > 0, then Q(x, y) is positive definite. If a < 0, then Q(x, y) is negative definite.
Proof. Notice that
4aQ(x, y) = 4a(ax2 + bxy + cy 2 ) = (2ax + by)2 − dy 2 .
(2.2)
Then if d < 0, the right-hand side of (2.2) is nonnegative and is equal to zero if and only if x, y = 0.
By considering the left-hand side of (2.2), we now see that if a > 0, Q(x, y) ≥ 0, and if a < 0,
Q(x, y) ≤ 0.
To further our understanding of quadratic forms, we now consider quadratic forms with a
specific type of discriminant.
Definition 2.12. A discriminant d of a binary quadratic form is called a fundamental discriminant
if d cannot be represented as d0 r2 with d0 ≡ 0, 1 (mod 4) and r > 1.
Example 2.13. For each of the forms Q0 , Q1 , and Q2 from our prior examples, the discriminant
is −23 and the coefficient of x2 is greater than zero. Thus each of these forms is positive definite.
Also, the discriminant −23 can only be factored as (−23)(±1)2 , so we also have that d = −23 is a
fundamental discriminant.
We note here that, as seen in the previous example, any discriminant that is square-free will be
a fundamental discriminant. We will be primarily concerned with quadratic forms with negative
fundamental discriminant, so we also note that the first ten negative fundamental discriminants
are -3, -4, -7, -8, -11, -15, -19, -20, -23, -24, and -31.
Next we state a definition that will help us to make some observations about the equivalence
of quadratic forms.
Definition 2.14. A primitive positive definite form ax2 + bxy + cy 2 is said to be reduced if
|b| ≤ a ≤ c, and b ≥ 0 if either |b| = a or a = c.
Example 2.15. We see that Q0 (x, y) = x2 + xy + 6y 2 satisfies the criteria to be a reduced primitive
positive definite form. However, the form Q(x, y) = 6x2 + xy + y 2 is not reduced.
7
The following theorem can help determine whether or not quadratic forms are properly equivalent.
Theorem 2.16. Every primitive positive definite form is properly equivalent to a unique reduced
form.
The proof of this theorem is omitted for brevity. (See [5], Theorem 2.8.)
Example 2.17. We can see that the forms Q0 (x, y) = x2 + xy + 6y 2 , Q1 (x, y) = 2x2 + xy + 3y 2 ,
and Q2 (x, y) = 2x2 − xy + 3y 2 are all reduced forms. We already saw that Q1 and Q2 are equivalent
but not properly equivalent. However, since each form is reduced, by Theorem 2.16, no two of these
three forms are properly equivalent.
We will end this section with the significant observation that the set C(d) of equivalence classes
of properly equivalent binary quadratic forms is an abelian group.
For brevity, we omit the proofs of the following lemma and theorem. (See [5], Lemma 3.2 and
Theorem 3.9.)
Lemma 2.18. Assume that Q(x, y) = ax2 + bxy + cy 2 and Q0 (x, y) = a0 x2 + b0 xy + c0 y 2 have
discriminant d and satisfy
gcd(a, a0 , (b + b0 )/2) = 1.
Then there is a unique integer B modulo 2aa0 such that
B ≡ b (mod 2a)
B ≡ b0
B2 ≡ D
(mod 2a0 )
(mod 4aa0 ).
We can now define a group structure on the set C(d).
Theorem 2.19. Let Q(x, y) = ax2 + bxy + cy 2 and Q0 x, y) = a0 x2 + b0 xy + c0 y 2 such that Q1 and
Q2 are both of discriminant d < 0, d ≡ 0, 1 (mod 4). Then
(Q1 ∗ Q2 )(x, y) = aa0 x2 + Bxy +
B2 − d 2
y ,
4aa0
where B is as described in Lemma 2.18. The set (C(d), ∗) forms an abelian group. The order of
the group C(d) is the number of equivalence classes, denoted h(d).
8
2.2
Quadratic Fields
There is an interesting connection between quadratic forms and quadratic fields that we will use.
To understand this connection, we begin by recalling basic definitions and theorems regarding
quadratic fields. Here we will assume a basic familiarity with algebraic number theory. For the
reader without a background in algebraic number theory, we recommend [16] for a nice introduction
to quadratic fields.
Recall, a field K is a number field if it is a finite degree field extension of Q. An element
α ∈ K is called an algebraic integer if the minimal polynomial of α is contained in Z[x]. The set of
algebraic integers OK forms a ring called the ring of integers of K.
√
A quadratic field is a degree two extension of Q. A √
quadratic field has the form Q( d) for
square-free d. Fix such a square-free d, and let K = Q( d). Then the ring of integers of K is
( [16], Theorem 3.2),
( √
if d 6≡ 1 (mod 4),
[1, d]
√
Od =
(2.3)
1+ d
[1, 2 ] if d ≡ 1 (mod 4),
where
[a, b] := {ma + nb : m, n ∈ Z} .
√
Example 2.20. In this paper, we will be primarily concerned with the quadratic field Q( −23).
The corresponding ring of algebraic integers is
√
1 + −23
O−23 = 1,
.
2
√
√
For future use, we recall the definitions of two norms. If α ∈ Q( d), with α = a + b d, then
the norm of α is defined to be
√
√
N (α) = (a + b d)(a − b d) = a2 − b2 d.
Recall, if α ∈ Od , N (α) ∈ Z (see [16], p. 49). We also note that if A is an ideal of OK , then the
norm of A is defined to be
N (A) = |OK /A|.
√
Additionally, we see that if (α) is a principal ideal where α = a + b d, then N ((α))=|a2 − b2 d|
(see [16], Corollary 5.10).
Remark 2.21. In order to reach our goals, we note that we spend most of this section simply
collecting the definitions and theorems from algebraic number theory that we will need in later
sections. We will omit many proofs of theorems; however, sources will be provided for each theorem
stated. To conclude the section, we prove Theorem 2.33.
Now we provide the definitions of the Legendre and Kronecker symbols along with propositions
about the ideals of Od .
9
Definition 2.22. Let a ∈ Z and p be an odd prime. The Legendre symbol is defined by

if a(p−1)/2 ≡ 1 (mod p),

1
a
= −1 if a(p−1)/2 ≡ −1 (mod p),

p

0
if a ≡ 0 (mod p).
Equivalently, in terms of quadratic residues,

if a is a quadratic residue modulo p and a 6≡ 0

1
a
= −1 if a is a quadratic non-residue modulo p,

p

0
if a ≡ 0 (mod p).
(mod p),
Definition 2.23. Let n be an integer with prime factorization n = upe11 ...pekk where u = ±1 and
each pi is prime. The Kronecker symbol is a function of integers a and n and is defined by
a
n
where for odd prime pi ,
a
pi
=
k ei
aY
a
u
i=1
pi
,
is the Legendre symbol, for pi = 2, we define

0
a 
= 1

2

−1
and we also define
if a is even,
if a ≡ ±1 (mod 8),
if a ≡ ±3 (mod 8),
a
= 1,
1
(
1
if a ≥ 0,
a
=
−1
−1 if a < 0,
and
(
1
=
0
0
a
if a = ±1,
otherwise.
As we will be considering the factorization of ideals, we state the following important theorem.
(See [16], Theorem 5.6.)
Theorem 2.24. Every non-zero ideal of Od can be written as a product of prime ideals uniquely
up to the order of the factors.
The following proposition is a useful fact about factors of principal ideals. (See [19], Chapter 2
Section 4.)
Proposition 2.25. Every ideal A of Od divides the principal ideal (α), where α = N (A).
10
In a√general number field, there is a method to compute the discriminant of the number field.
For Q( d), the discriminant Dd is given by ( [16], Section 3.1)
(
d if d ≡ 1 (mod 4),
Dd =
4d if d ≡ 2, 3 (mod 4).
Next we will consider the factorization of (p) in Od . (See [13], Section 6.2 and [8] Proposition
13.1.4.)
Proposition 2.26. Let p be any prime. Then in Od we have

0
0 if Dd = 1,

P
P
where
P
=
6
P

p


Dd
(p) = P
if p = −1,


 2

P
if Dpd = 0,
where P, P 0 are prime ideals of Od .
√
5−1
Example 2.27.
Consider the quadratic field Q( −23). In this case, −23 2 = −529 ≡ 1 (mod 5)
and thus −23
= 1, so we see that (5) is a prime ideal in O−23 . Similarly, we also know that
5
2
(23) = P for a prime ideal P because −23
= 0.
23
Next we introduce the idea of fractional ideals of a quadratic field.
√
√
Definition 2.28. A subset F ⊂ Q( d) is called a fractional ideal of Q( √d) if there exists β ∈ Od ,
β 6= 0, such that βF is an ideal of Od . The set of fractional ideals of Q( d) is denoted Fd .
Notice that if F is a fractional ideal, then
F =
α
:α∈A
β
for some ideal A of Od and element β ∈ Od . Then βF = A. We define the multiplication of two
fractional ideals F1 and F2 by
F1 F2 = (β1 β2 )−1 A1 A2 ,
where A1 and A2 are the ideals of Od such that F1 = β1−1 A1 and F2 = β2−1 A2 .
Next we consider the group structure of the set Fd , (see [16], Theorem 5.5.)
√
Theorem 2.29. The set Fd of nonzero fractional ideals of Q( d) forms an abelian group under
multiplication. The identity of the group is Od . The inverse of the fractional ideal F is
n
o
√
F −1 = α ∈ Q( d) : αF ⊆ Od .
11
We also note that the set of principal ideals Bd ⊂ Fd , is a subgroup of Fd .
√
Definition 2.30. The quotient group Hd√= Fd /Bd is called the ideal class group of Q( d). The
order of Hd is hd , the class number of Q( d).
Example 2.31. If hd = 1, then all fractional ideals are in the same equivalence class in Hd , so
each fractional ideal is equivalent to the principal ideal (1) = Od modulo multiplication by principal
ideals. That
is, for each fractional ideal A there exists α ∈ Od such that (α)A = (1) = Od . Then
A = α1 , and we can see that every fractional ideal, and thus every ideal, is principal.
It is important to note that regardless of the class number, the principal ideals form one equivalence
class within the class group Hd . Additionally, it is worth noting that each ideal class contains an
actual ideal that is not fractional. To see this, notice that if F = β −1 A for some ideal A and
β ∈ Od , then F (β) = A will be an integral ideal in the ideal class F Bd . Such an ideal of Od that is
not fractional is called an integral ideal.
We are now ready to see the connection between binary quadratic forms and quadratic fields
as found in [5], (see Theorem 7.7).
Theorem 2.32. Let d < 0 be a fundamental discriminant. Then the map φ : C(d) → Hd defined
by
"
#
√
−b + b2 − 4ac
2
2
φ(ax + bxy + cy ) = a,
2a
is a group isomorphism between
the class group of binary quadratic forms of discriminant d and
√
the ideal class group of Q( d). In particular, h(d) = hd .
We next see that there is an even deeper connection between binary quadratic forms and ideal
classes.
Theorem 2.33. For a positive-definite binary quadratic form Q with fundamental discriminant
d < 0,
r(Q, n) = wd · r(A, n),
where wd is the number of units in Od , and r(A, n) is the number of integral ideals with norm n in
the corresponding ideal class A.
The rest of this section we build toward proving Theorem 2.33. First we compute the number of
units in the ring of integers Od .
Theorem 2.34. The number of units in Od is

 4 for d = −1,
6 for d = −3,
wd =

2 for d = −2 or d < −4.
12
(2.4)
√
Proof. Let d < 0, and let α = a + b d ∈ Od be a unit. Then there exists β ∈ Od such that αβ = 1.
In this case, 1 = N (1) = N (αβ) = N (α)N (β). Since α, β ∈ Od , N (α), N (β) ∈ Z. Thus, we must
have N (α) = ±1. This happens exactly when
N (α) = a2 − db2 = 1
(2.5)
since a2 − db2 > 0 when d < 0.
By (2.3), if d ≡ 2, 3 (mod 4), then a, b ∈ Z. Then by (2.5), a = ±1 and b = 0, or, if d = −1,
also a = 0 and b = ±1.
By (2.3), if d ≡ 1 (mod 4), then (2a)2 − d(2b)2 = 4 where 2a, 2b ∈ Z. Then by (2.5), either
a = 1 and b = 0, or, if d = −3, also a = ±1/2 and b = ±1/2.
Counting the number of units for each case, we see that the claim of the theorem holds.
Recall that two forms Q0 (x, y) and Q1 (x, y) are properly equivalent when
Q0 (x, y) = Q1 A(x, y) = Q1 (px + qy, rx + sy),
where A = ( pr qs ) ∈ SL2 (Z). If for A ∈ SL2 (Z) we have that Q(x, y) = QA(x, y), we call A an
automorph of the quadratic form Q. It will be useful to know the number of automorphs for a
quadratic form. (For the following theorem, see [20], Section 8.)
Theorem 2.35. If Q(x, y) = ax2 + bxy + cy 2
then the number of automorphs for Q is

 6
4
aQ =

2
is a primitive quadratic form with discriminant d,
for d = −3,
for d = −4,
for d < −4.
Moreover, the set of automorphs for Q can be described as




 t−bu
(±2, 0) or (±1, ±1) if d = −3,
−cu
2
: (t, u) = (±2, 0) or (0, ±1) if d = −4,
t+bu


 au

2
(±2, 0) if d < −4.
(2.6)





We note that by comparing (2.4) and (2.6) for
quadratic form Q with fundamental
√
√ each primitive
discriminant d < 0 and by observing that Q( −4) = Q( −1), it can be seen that aQ = wd .
We are nearly able to prove Theorem 2.33, but first we state the following lemmas, (see [5],
Section 7B).
Lemma 2.36. If Q(x, y) = ax2 + bxy + cy 2 is a positive-definite quadratic
form with fundamental
√
−b+ d
discriminant d, then a[1, τ ] is an integral ideal of [1, aτ ], where τ = 2a . Additionally,
Od = [1, aτ ].
13
Lemma 2.37. Let A be the ideal class in Hd corresponding to the positive definite quadratic form
Q(x, y) = ax2 + bxy + cy 2 with fundamental discriminant d < 0 via the isomorphism defined
in Theorem 2.32. Then every integral ideal in A can be expressed as α[1, τ ] for some α ∈ Od .
Additionally, if A = α[1, τ ] is an integral ideal of Od , then N (A) = N (α)/a.
We are now ready to prove Theorem 2.33.
Proof. Let Q(x, y) = ax2 + bxy + cy 2 be a positive definite binary quadratic form with fundamental
discriminant d = b2 − 4ac < 0. As in Theorem 2.32, the quadratic form Q(x,√y) corresponds
d
to the ideal class A, containing the fractional ideal A = a[1, τ ], where τ = −b+
2a . By Lemma
2.36, we know that A is an integral ideal of Od . Let B be an integral ideal of Od of norm n in
A. By Lemma 2.37, we know that B = α[1, τ ] for some α ∈ Od and that N (α)/a = n. Since
α, ατ ∈ Od = [1, aτ ], we know α = p + qaτ and ατ = r + saτ for some p, q, r, s ∈ Z. Then using
the fact that (p + qaτ )τ = r + saτ and the fact that, by the definition of τ , aτ 2 = −bτ − c, we see
that p = as + bq.
Now we can see that
N (α)
n=
a
=
1
N
a
=
1
a
√ !
qb q d
p−
+
2
2
√ !
√ !
qb q d
qb q d
p−
+
p−
−
2
2
2
2
1 2
(p − bpq + acq 2 )
a
1
((as + bq)2 + absq + acq 2 )
=
a
= as2 + bsq + cq 2
=
= Q(s, q)
Then we see that Q represents n. Next notice that if B 6= B 0 ∈ A with N (B) = n = N (B 0 ),
then B = α[1, τ ] and B 0 = α0 [1, τ ] with α 6= α0 . Using B 0 in the process above, we find numbers
s0 , q 0 ∈ Z such that Q(s0 , q 0 ) = n. Also, we must have that (s0 , q 0 ) 6= (s, q) or else we would have
α = α0 . Then we have that each integral ideal of norm n corresponds to a solution of Q(x, y) = n,
and the solutions are different if the ideals are not equal.
Next we consider any (s, q) ∈ Z2 with Q(s, q) = n. By verifying closure under multiplication
by elements in Od using straightforward calculations, it can be shown that if p = as + bq and
α = p + qaτ , then B = α[1, τ ] is an integral ideal in A. Then every solution of Q(x, y) = n
corresponds to an integral ideal in A.
So far we have seen that each integral ideal of norm n in A corresponds to a different solution
of Q(x, y) = n, and every solution of Q(x, y) = n corresponds to at least one ideal of norm n
14
in A. In terms of r(Q, n) and r(A, n), we now have that r(Q, n) ≥ r(A, n). What remains to
be seen is that, in fact, each integral ideal in A leads to exactly wd solutions of Q(x, y) = n,
where wd is the number of units in the ring of integers Od . To see this, using basic calculations
along with the explicit statement of the automorphs of Q in Theorem 2.35, it can be shown that if
Q(x, y) = Q(tx + uy, vx + wy) (that is, if ( vt wu ) is an automorph for Q), then the solution (s, q) with
Q(s, q) = n and the solution (tx + uy, vx + wy) correspond to the same ideal in C. For example,
if (s, q) is a solution that corresponds to B = α[1, τ ], it is relatively easy to see the the solution
(−s, −q) corresponds to −α[1, τ ] = B as well.
2.3
Modular Forms
We now brush the surface of the basics of modular forms. We primarily rely on [6] and [14] as
references as we provide definitions and theorems.
Definition 2.38. The upper-half plane H is the set of complex numbers
H = {z ∈ C : Im(z) > 0} .
Definition 2.39. Let f be a function, f : H → C, and let k be an integer. The function f is called
a modular form of weight k on SL2 (Z) if
1. for all z ∈ H and
a b
c d
∈ SL2 (Z),
az + b
f
= (cz + d)k f (z).
cz + d
(2.7)
2. f is holomorphic on H.
3. f is holomorphic at i∞.
Example 2.40. We will show that if f is a modular form of weight k and g is a modular form
of weight l, then f g is a modular form of weight k + l. First note that the product f g will still be
holomorphic on H and at i∞. Additionally we see
az + b
az + b
az + b
(f g)
= f
g
cz + d
cz + d
cz + d
= (cz + d)k f (z)(cz + d)l g(z)
= (cz + d)k+l (f g)(z).
Modular forms can also be considered as functions on lattices of C, so we will take some time
to develop an understanding of lattices.
15
Definition 2.41. If ω1 , ω2 ∈ Z with ω1 /ω2 ∈
/ R, then the lattice of C generated by ω1 and ω2 is
defined to be
Λ =< ω1 , ω2 >= {mω1 + nω2 : m, n ∈ Z} .
We call {ω1 , ω2 } a basis of Λ.
We note that each lattice is an additive subgroup of C. If Λ0 =< α1 , α2 > is a subgroup of
Λ =< ω1 , ω2 >, then we know that each αi is an integral combination of ω1 and ω2 . Then we have
α1
ω1
=A
,
(2.8)
α2
ω2
for some 2 × 2 integral matrix A. We call Λ0 a sublattice of Λ if α1 /α2 ∈
/ R. In fact, letting A be
as in (2.8), Λ0 := Λ(A) is a sublattice exactly when det(A) 6= 0. We note that every 2 × 2 integral
matrix A with nonzero determinant corresponds to exactly one sublattice Λ(A) of Λ. Additionally,
if Λ(A) = Λ(B), then A = U B for a matrix U with determinant ±1. We say the index of the
sublattice Λ0 of Λ is | det(A)| where Λ0 = Λ(A). For a more detailed discussion of sublattices, we
refer the reader to [17].
An additional proposition about sublattices will be useful in a later section.
Proposition 2.42. The sublattices of Λ of index m are in bijective correspondence with the 2 × 2
integral matrices
a
b
,
0
d
such that 0 < b < d and ad = m.
We omit the proof of this proposition for brevity, (see [17]).
Now we are prepared to continue our discussion of modular forms. A function F mapping from
the set of lattices of C to C is a modular form of degree −k if F (λΛ) = λ−k F (Λ) for all lattices Λ
of C and all λ ∈ C with λ 6= 0. We can see the connection between a modular form F on lattices
and and a modular form f on H by letting f (z) = F (< z, 1 >). Then as a result,
F (Λ) = ω2−k f (ω1 /ω2 ),
where {ω1 , ω2 } is any oriented basis of Λ. Note that we can choose an ordering ω1 , ω2 such that
ω1 /ω2 ∈ H.
Example 2.43. If f is a modular form of weight k,
−1
−1
f
=F
1,
= z k F (< z, −1 >) = z k F (< z, 1 >) = z k f (z).
z
z
We note that this corresponds to (2.7) for the matrix 01 −1
0 .
16
It is sometimes convenient to view a modular form in terms of its Fourier expansion at i∞.
For more details about Fourier expansions, see [6], Section 1.1. A modular form f has the Fourier
expansion at i∞
∞
X
f (z) =
an q n ,
n=0
where the an are called the Fourier coefficients of f , and q = e2πiz . If a0 = 0, then f is a called a
cusp form. Note that this means that f vanishes at i∞.
Example 2.44. We define the ∆-function by
∆(z) = q
∞
Y
(1 − q n )24 ,
n=1
where q = e2πiz . The ∆-function is a cusp form of weight 12 on SL2 (Z). (See [4], p. 20.)
Our next topic will be modular forms on congruence subgroups.
Definition 2.45. If N is a positive integer, then define the level N congruence subgroups Γ0 (N )
and Γ1 (N ) by
a b
∈ SL2 (Z) : c ≡ 0 (mod N )
Γ0 (N ) =
c d
and
Γ1 (N ) =
a b
c d
∈ SL2 (Z) : a ≡ d ≡ 1
(mod N ), and c ≡ 0
(mod N ) .
Definition 2.46. Let f be a function, f : H → C, and let k be an integer. The function f is said
to be a modular form of weight k on the congruence subgroup Γ of level N if
1. for all z ∈ H and
a b
c d
∈ Γ,
f
az + b
cz + d
= (cz + d)k f (z).
2. f is holomorphic on H.
3. if A = ac db ∈ SL2 (Z), then
−k
g(z) = (cz + d)
f
az + b
cz + d
has a Fourier expansion of the form
g(z) =
X
n≥nA ≥0
where qN = e2πiz/N and aA (nA ) 6= 0.
17
n
aA (n)qN
,
If nA > 0 for each A ∈ SL2 (Z), then f is called a cusp form on Γ.
A modular form f on congruence subgroup Γ has the Fourier expansion at i∞
f (z) =
∞
X
an q n ,
n=0
where q = e2πiz . (See [14], Chapter 1.)
We denote the set of modular forms of weight k by Mk and the set of modular forms of weight
k on a congruence subgroup Γ by Mk (Γ). Note that Mk = Mk (Γ0 (1)). Similarly, we denote the
set of cusp forms of weight k by Sk and the set of cusp forms of weight k on a congruence subgroup
Γ by Sk (Γ). Each of the sets Mk (Γ) and Sk (Γ) are complex vector spaces. We will briefly examine
the structure of Mk and Sk as a vector space over C with the operations of function addition and
scalar multiplication of functions by elements of C. It is fairly straightforward to verify that the
properties of a vector space are satisfied by each of these sets. As an example of the necessary
calculations,
we will show that Mk and Sk are closed under addition. For f, g ∈ Mk , we have that
for all ac db ∈ SL2 (Z),
az + b
az + b
az + b
= f
+g
(f + g)
cz + d
cz + d
cz + d
= (cz + d)k f (z) + (cz + d)k g(z)
= (cz + d)k (f + g)(z).
When examining the cusp forms, if f, g ∈ Sk ⊂ Mk , we already know that f + g ∈ Mk . Then
it remains to show that the constant term of Fourier expansion of f + g is zero. However, since
both of the Fourier expansions for f and g have constant term zero, it is seen by adding these two
expansions together that the Fourier expansion for f + g will also have constant term zero. The
rest of the verification of vector space structure will be left to the reader.
We now turn our focus to Dirichlet characters modulo n.
Definition 2.47. A character on a finite abelian group G is a group homomorphism from G to C∗ .
A Dirichlet character modulo n, χ : Z → C∗ , is a character on the group (Z/nZ)∗ that additionally
satisfies
χ(m) = 0 if gcd(m, n) 6= 1
and
χ(m) = χ(m + n) for all m ∈ Z.
Example 2.48. There are two Dirichlet characters modulo 3. We can see that
(
1 if n ≡ 1, 2 (mod 3)
χ0 (n) =
0 if n ≡ 0 (mod 3).
18
and


1
χ1 (n) = −1


0
if n ≡ 1 (mod 3)
if n ≡ 2 (mod 3)
if n ≡ 0 (mod 3).
are the Dirichlet characters modulo 3.
If χ is a Dirichlet character modulo N, then we say that a modular
form f ∈ Mk (Γ1 (N )) (resp.
Sk (Γ1 (N ))) has Nebentypus character χ if for all z ∈ H and ac db ∈ Γ0 (N ),
az + b
= χ(d)(cz + d)k f (z).
f
cz + d
The space of modular forms (resp. cusp forms) of weight k on Γ0 (N ) with Nebentypus character χ
is denoted by Mk (Γ0 (N ), χ) (resp. Sk (Γ0 (N ), χ)).
Modular forms can also be defined to have half-integral weight. (See [14], p. 10.) An important
example is Dedekind’s eta-function which is denoted by η(z). Dedekind’s eta-function is defined by
η(z) := q 1/24
∞
Y
(1 − q n )
(2.9)
n=1
where q = e2πiz . Notice that ∆(z) = η(z)24 . More about Dedekind’s eta-function can be found
in [14].
We will not venture into the theory of half-integral weight modular forms because the theory will
not be necessary for our purposes. However, we will see the η-function play an important role in
the rest of our discussion.
While we have seen the basic definitions needed for our purposes, the reader can find these
definitions and a more detailed discussion of modular forms in [6] or [14].
19
Chapter 3
Relating Quadratic Forms and
Modular Forms
3.1
Theta Series
In this section, we focus on the subject of theta series for quadratic forms. Since we are only
interested in binary quadratic forms, we will consider only this case. This theory, along with the
generalization to all quadratic forms, is described by Zagier in [4].
Let Q(x, y) = ax2 + bxy + cy 2 be a positive definite quadratic form.
Definition 3.1. The theta series of Q is defined to be
θQ (z) =
∞
X
r(Q, n)q n ,
n=0
where q = e2πiz , and r(Q, n) denotes the number of representations of n by Q.
We will soon see that θQ (z) is a modular form, but first we must gather more information. The
quadratic form Q(x, y) can be written in the form
1
Q(x, y) = (x, y)A(x, y)t ,
2
(3.1)
b . Note that A is symmetric. The symmetric matrix A is called positive definite
where A = 2a
b 2c
if (x, y)A(x, y)t > 0 for all nonzero (x, y) ∈ R2 . Then we can see that since Q is positive definite,
we must have that A is positive definite.
Proposition 3.2. Every positive definite matrix A is invertible.
20
We omit the proof of this proposition for brevity. (See [11], Chapter 6 Section 7.)
By Proposition 3.2, we may now assume that A from (3.1) has an inverse, and we will use this fact
in the next definition.
Definition 3.3. If Q(x, y) = 12 (x, y)A(x, y)t , then the level of Q is the smallest positive integer,
N = NQ , such that N A−1 is again a matrix with integral elements and the aii are even for each i.
Example 3.4. If Q0(x, y) = x2 + xy + 6y 2 , then Q0 (x, y) = (x, y)A(x, y)t where A = ( 21
12/23 −1/23
we have that A−1 = −1/23 2/23 . From this, we can see that the level of Q0 is 23.
1
12 ).
Then
Definition 3.5. If Q(x) = 21 (x, y)A(x, y)t , then the discriminant ∆ = ∆Q of A is defined to be
det(A).
b
Example 3.6. Let Q(x, y) = a2 + bxy + cy 2 . Then A = 2a
b 2c , and the discriminant of A is
det(A) = b2 − 4ac. Then we see that the discriminant of A is the discriminant of Q.
There is an associated character to our quadratic form Q(x, y) = (x, y)A(x, y)t with level N and
discriminant ∆ of A. The associated character, χ∆ (Kronecker symbol), is the unique Dirichlet
character modulo N satisfying
∆
χ∆ (p) =
(Legendre symbol)
p
for any odd prime p that does not divide N . (See [15], p. 303.)
These ideas lead us to an important theorem.
Theorem 3.7. Let Q be a positive definite binary quadratic form of level N and discriminant ∆.
Then θQ is a modular form on Γ0 (N ) of weight 1 and character χ∆ . In particular,
az + b
θQ
= χ∆ (a)(cz + d)k θQ (z)
cz + d
for all z ∈ H and ac db ∈ Γ0 (N ).
We omit the proof of this theorem for brevity. (See [1], Theorem 2.2.)
3.1.1
Hecke Theory
Let Λ be a lattice on C with basis {ω1 , ω2 }. For each integer m ≥ 1, for any given k there exists a
linear operator Tm : Mk → Mk , called the mth Hecke operator. For a modular form F of degree −k
on lattices Λ ⊂ C, we have (up to a suitable normalizing constant ensuring that P
the image of a form
with integral Fourier coefficients has integral Fourier coefficients) Tm F (Λ) =
F (Λ0 ) where the
21
sum is over sublattices Λ0 ⊂ Λ of index m. Translating this using the fact that F (< 1, z >) = f (z)
for a modular form f of weight k, the operation is
X
az + b
k−1
−k
(3.2)
Tm f (z) = m
(cz + d) f
cz + d
a b ∈SL2 (Z)\Mm
c d
for z ∈ H, where Mm denotes the set of 2 × 2 integral matrices of determinant m, and SL2 (Z)\Mm
is a set of representatives from Mm such that each sublattice Λ0 = Λ(A) for exactly one element
A ∈ SL2 (Z)\Mm . (Note that the constant mk−1 has been introduced for later convenience.)
Our current goal is to use our theory to express Tm f (z) in a more useful
form. By Proposition
a
b
2.42, we may choose our set SL2 (Z)\Mm to be the set of matrices 0 d such that 0 < b < d and
ad = m. Combining this fact with (3.2), we have
X
k−1
Tm f (z) = m
−k
d
f
ad=m
a,d>0
0<b<d
az + b
d
,
or, in a another form,
Tm f (z) = m
X
k−1
X
−k
d
f
b mod d
ad=m
a,d>0
az + b
d
.
Using the Fourier expansion of f , we can now carry this equation a few steps forward:
X
X X
Tm f (z) = mk−1
d−k
an e2πin(az+b)/d
b mod d n≥0
ad=m
a,d>0
X
= mk−1
d−k
X
n≥0
ad=m
X
an e2πinaz/d
e2πibn/d
b mod d
a,d>0
= mk−1
X
d−k d
X m k−1 X
d|m
d>0
d
2
an q mn/d .
n≥0
d|n
We can conclude that


Tm f (z) =
an e2πinaz/d
n≥0
d|n
ad=m
a,d>0
=
X
X X
 n
k−1

r
a
2 q .
mn/r


n≥0
r|(m,n)
r>0
22
(3.3)
Since Tm is a linear operator on functions, it is natural to consider the eigenfunctions of Tm . We
call f ∈ Mk a simultaneous eigenform if it is an eigenfunction of the Tm for all m with corresponding
eigenvalues λm . If f is a simultaneous eigenform, then
Tm f (z) = λm f (z) = λm
∞
X
an q n .
(3.4)
n=0
We have the the coefficient of q in the expansion of Tm f (z) is am when using (3.3) and λm a1
when using (3.4). Thus, λm a1 = am . We can now conclude that if f is not identically zero, then
a1 6= 0. If we normalize f by dividing all coefficients by a1 , then f is called a normalized Hecke
eigenform. In this case we have that
Tm f = am f
(3.5)
and
am an =
X
rk−1 amn/r2
if m, n ≥ 1.
(3.6)
r|(m,n)
The following is a theorem of Hecke.
Theorem 3.8. Mk has a basis of normalized simultaneous eigenforms for all k, and that basis is
unique.
We omit the proof of this theorem for brevity. (See [9], Section 7.3.)
We conclude this section by noting that Hecke’s theory generalizes to congruence subgroups of
SL2 (Z). (See [4], p. 39).
3.1.2
L-series of Eigenforms
We begin by defining the L-series of a modular form.
P
n
Definition 3.9. If f (z) = ∞
n=0 an q is a modular form (resp. a normalized simultaneous eigenform) in any space of modular forms, then we define the L-series (resp. Hecke L-series) of f to
be
∞
X
an
L(f, s) =
.
ns
n=1
The goal of this section is to develop a different and useful representation for L(f, s) where f
is a normalized simultaneous eigenform in Mk . In particular, we will show that
L(f, s) =
Y
p∈P
1 − ap
23
1
+ pk−1−2s
p−s
where P is the set of prime numbers. First, using (3.6) when (m, n) = 1, we obtain that
am an = amn .
(3.7)
Using (3.6) when m = pv and n = p for p prime, we obtain that
apv+1 = ap apv − pk−1 apv−1 .
(3.8)
By (3.7), the Fourier coefficients of f are multiplicative. Thus, (see [4], Page 39), the Hecke
L-series of f has the Euler product,
Y
ap ap2
(3.9)
L(f, s) =
1 + s + 2s + ... ,
p
p
p∈P
where P is the set of prime numbers.
Next we will use a proposition to rewrite the Euler product in an improved form.
P
n
Proposition 3.10. Let f (z) = ∞
n=0 an q be a normalized simultaneous eigenform in Mk . Then
for each prime p,
∞
X
1
apv p−vs =
.
−s
1 − ap p + pk−1−2s
v=0
Proof. First, define a new series
∞
X
!
apv p−vs
(1 − ap p−s + pk−1−2s ).
v=0
The coefficient of p−s in this series is ap − a1 ap = ap − ap 0 since a1 = 1 for a normalized eigenform.
For n ≥ 1, the coefficient of p−(n+1)s in this series is apn+1 − apn ap + apn−1 pk−1 = 0 by (3.8). Then
we find that the sum of the series is a1 = 1.
Applying Proposition 3.10 to (3.9), we can see that
L(f, s) =
Y
p∈P
1 − ap
1
,
+ pk−1−2s
p−s
which is Hecke’s fundamental Euler product for the L-series of a normalized Hecke eigenform
f ∈ Mk .
24
3.2
Modular Forms and Algebraic Number Theory
We can learn more about the number of representations r(Q, n) of n by the binaryP
quadratic form
∞
n
Q(x, y) = ax2 + bxy + cy 2 by considering the weight one theta series θQ (z) =
n=0 r(Q, n)q
from Definition 3.1. In this section, consider the case where Q is a binary quadratic form with
fundamental discriminant d < 0.
Recall that by Theorem 2.33, we know r(Q, n) = wd · r(A, n). Then we can express the L-series
L(θQ , s) =
∞
X
r(Q, n)
n=1
ns
as
L(θQ , s) = wd
X
N (A)−s ,
A∈A
where the sum is over the integral ideals of A. We will define the “partial zeta-function” ζK,A (s)
by the following:
X
ζK,A (s) =
N (A)−s .
A∈A
√
Let χ be a character on Hd , and let K = Q( d). Then we can define a new L-series
LK (s, χ) =
X χ(A)
,
N (A)s
A
where the sum is over the integral ideals of Od . This can be rewritten as
X
X
X
LK (s, χ) =
χ(A)
N (A)−s =
χ(A)ζK,A (s).
A∈Hd
A∈Hd
A∈A
Next we will see a connection
forms and LK (s, χ) thus justifying our L-series
P between modular
n . Then if we define
notation. We know θQ (z) = ∞
r(Q,
n)q
n=1
θA (z) = wd
∞
X
r(A, n)q n ,
n=1
we have that θQ (z) = θA (z) where A is the ideal class corresponding to Q. Next we introduce the
function
X
fχ (z) = wd−1
χ(A)θA (z).
(3.10)
A∈Hd
Since θA (z) = θQ (z), using Theorem 3.7 along with the fact that adding weight one modular forms
results in a weight one modular form, we can conclude that fχ (z) is a weight one modular form.
25
Also, using Definition 3.9, it can be seen that the L-series of fχ (z) is LK (s, χ). Indeed, we have
∞ P
X
A∈Hd χ(A)r(A, n)
L(fχ , s) =
ns
n=1
X
=
χ(A)
A∈Hd
X
=
n=1
χ(A)
A∈Hd
X
=
∞
X
r(A, n)
X
ns
N (A)−s
A∈A
χ(A)ζK,A (s)
A∈Hd
= LK (s, χ)
By the unique prime decomposition of ideals in K, we can use a process to sieve the prime
ideals out of the sum and see that LK (s, χ) has a resulting Euler product. (See [10], Section 1.5.)
The condition of f being a Hecke eigenform is equivalent to its L-function series having an Euler
product, (see [6]). Then we can conclude that fχ is a Hecke eigenform. The following theorem
shows us what form the the Euler product will have.
P
n
Theorem 3.11. Let f ∈ Mk (Γ0 (N ), χ), f (z) = ∞
n=0 an q . The following are equivalent:
1. f is a normalized eigenform
2. L(s, f ) has an Euler product expansion
L(s, f ) =
Y
(1 − ap p−s + χ(p)pk−1−2s )−1 ,
p
where the product is taken over all primes.
We omit the proof of this theorem for brevity. (See [6], Theorem 5.9.2.)
Next we will consider the specific character χ = χ0 where χ0 is the trivial character.
√
Definition 3.12. Let K = Q( d). Then the Dedekind zeta function of K is
X
ζK (s) =
N (A)−s ,
A
where the sum is over the integral ideals of Od .
If χ = χ0 , then we can see that
LK (s, χ) =
X
N (A)−s = ζK (s).
A
26
As we progress, we will rely heavily on the character
Dd
(Kronecker symbol).
Dd (n) =
n
Our goal now is to prove the following theorem.
Theorem 3.13. Let K be a quadratic field. Then for Re(s) > 1,
ζK (s) = ζ(s)L(s, Dd ),
P∞
−s is the Riemann zeta function and L(s, ) =
where ζ(s) =
Dd
n=1 n
Dirichlet L-series of the character Dd (n).
P∞
n=1 Dd (n)n
−s
is the
We will use this theorem to gain more information about the number r(Q, n). In particular, we
will prove the following proposition.
Proposition 3.14. Let Q be a positive definite binary quadratic form with fundamental discriminant d < 0. Then
X
X
r(Q, n) = wd
Dd (b).
[Q]∈C(d)
b|n
Remark 3.15. In order to reach our goals, we will omit the proofs for the lemmas leading to the
proof of Theorem 3.13. For more details regarding the following lemmas, we refer the reader to [19]
where the following lemmas are given with more detail. We will also provide additional sources for
lemmas as we proceed.
Lemma
3.16.
Let (an ) be a sequence of complex numbers. Suppose there exist c, r > 0 such that
P
P∞
M
n=1 an ≤ cM r for all M ≥ 1. Then the Dirichlet series n=1 an n−s converges for all s with
Re(s) > r.
(For proof, see [12], Chapter 7 Lemma 1.)
Lemma 3.17. Let (an ) be a multiplicative sequence of complex numbers.
Suppose there
exists c > 0
PM
P∞
Q P∞
−s
−js
such that n=1 |an | ≤ cM for all M ≥ 1. Then n=1 an n = p
for all s with
j=0 apj p
Re(s) > 0.
(For proof, see [2], Theorem 11.7.)
Let υm be the number of ideals of Od with norm m.
Lemma 3.18. The sequence (υn ) is multiplicative. Additionally, for prime p and Re(s) > 1,

−s )−2 if Dd = 1,

(1
−
p


∞

p X
υpn
−s
−1
= (1 − p ) if Dpd = 0,

pns

n=0

(1 − p−2s )−1 if Dd = −1.
p
27
(For proof, see [19], Lemma 5.1.)
P
Now we are ready to consider specifically ζK (s). First we note that since ζK
= A N (A)−s
P(s)
−s
and υn is the number of ideals of Od with norm n, then we have that ζK (s) = ∞
n=1 υn n .
Lemma 3.19. Let
( √
√
Ad =
−d −d ≡ 1, 2 (mod 4),
−d ≡ 3 (mod 4).
−d
2
There exists a constant c such that
M
X
√
hd π
υ
−
M
≤ c M,
n
Ad wd n=1
for all M ≥ 1.
(For proof, see [19], Proposition 5.2.)
This brings us to a point where we can prove Theorem 3.13.
Proof. To show that ζK (s) converges, first we will show that there exists a constant c such that
M
X
hd π +c M
(3.11)
υn ≤ Aw n=1
for all M ≥ 1. By Lemma 3.19, we know there exists a constant c such that
M
X
√
hd π
υn −
M ≤ c M
Ad wd n=1
for all M ≥ 1. Using the reverse triangle inequality, we see
X
M
X
h π
√
M
h
π
d
d
M ≤ M ≤ c M.
υn − υn −
Ad wd
Ad wd n=1
Then
n=1
M
h π
X
√
d
υn − M ≤ c M ,
Ad wd
n=1
and we can conclude that
M
X
hd π υn ≤ +c M
Aw n=1
for all M ≥ 1. Notice that by Lemma 3.16, using (3.11), we know that ζK (s) converges for
Re(s) > 1.
28
Next we will show that ζK (s) has the Euler product
Y
Y
ζK (s) =
(1 − p−s )−2
(1 − p−s )−1
D
p, pd =1
p,
Dd
p
Y
p,
=0
Dd
p
(1 − p−2s )−1 .
=−1
Since (υn ) is multiplicative and υn > 0 for each n, by Lemma 3.17,


∞
∞
Y X
X

υn n−s =
υpj p−js .
p
n=1
Then by Lemma 3.18,


∞
Y X

υpj p−js  =
p
j=0
Y
j=0
Y
(1 − p−s )−2
D
p, pd =1
p,
Dd
p
Y
(1 − p−s )−1
p,
=0
Dd
p
(1 − p−2s )−1 .
=−1
Thus, we have
Y
ζK (s) =
Y
(1 − p−s )−2
D
p, pd =1
p,
Dd
p
Y
(1 − p−s )−1
=0
p,
Dd
p
(1 − p−2s )−1 ,
(3.12)
=−1
where the first product is over the set of primes, p, such that Dpd = 1 and similarly for the other
two products. Continuing to simplify (3.12), we see
Y
Y
Y
ζK (s) =
(1 − p−s )−2
(1 − p−s )−1
(1 − p−s )−1 (1 + p−s )−1
p,
=
Dd
p
Y
=1
p,
=
Y
p,
(1 − p−s )−1
p
Dd
p
Y
(1 − p−s )−1
p
Dd
p
Y
=0
p,
p,
1−
p
Dd
p
p−s
Dd
p
Y
(1 − p−s )−1
=1
Dd
p
=−1
(1 + p−s )−1
=−1
−1
= ζ(s)L(s, K ).
Using Theorem 3.13, we will now prove Proposition 3.14.
Proof. Recall that L(θQ , s) =
P∞
r(Q,n)
ns
= wd ζK,A (s). Then we have that for Re(s) > 1,
X
ζK (s) =
ζK,A (s)
n=1
A∈Hd
=
=
1
wd
X
∞
X
r(Q, n)
ns
[Q]∈C(d) n=1
P
∞
1 X [Q] r(Q, n)
.
wd
ns
n=1
29
(3.13)
Note that we are able to interchange sums since ζK (s) converges absolutely on some half-plane. By
Theorem 3.13,
∞
∞ X
X
1 X Dd 1
ζK,A (s) =
.
(3.14)
ns
n ns
A∈Hd
n=1
n=1
By comparing coefficients of n−s , we will see that for each n,
X
X
r(Q, n) = wd
Dd (b).
[Q]
b|n
P
The coefficient of n−s in (3.13) is w1d [Q] r(Q, n). In (3.14), n−s appears exactly when the product
of the indices of the two sums is n. In such a case, when jk = n, we see that we collect the term
Dd
−s Dd
−s
j
k =
n−s .
k
k
P
From this we can see that the coefficient of n−s in (3.14) is b|n Dd (b).
With the theory we have developed, we wish to return to our consideration of the weight one
modular form fχ (z) from (3.10). We will continue to let χ = χ0 . Using our theory, we now see
fχ0 (z) =
=
=
=
=
1 X
θA (z)
wd
A
1 X
θQ (z)
wd
[Q]


∞
X
X
1

r(Q, n) q n
wd
n=0
[Q]


∞
1 X  X Dd  n
1 X
r(Q, 0) +
wd
q
wd
wd
b
n=1
[Q]
b|n


∞
h(d) X X Dd  n
+
q .
wd
b
n=1
(3.15)
(3.16)
(3.17)
b|n
Using the fact the θQ (z) = θA (z) when [Q] and A are corresponding classes, it can be seen that
for any character χ,
X
fχ (z) = wd−1
χ(A)θQA (z),
A
where QA is a quadratic form from the class [Q] corresponding to A. Additionally, if the order of
χ is greater than two, then fχ is a cusp form (see [4], p.42).
30
Chapter 4
The Case d = −23
We now can consider the case d = −23 as given by Zagier in [4]. In the case of d = −23, it turns
out that hd = 3. (See [3], p. 42.) Then we know by Theorem 2.32 that there exist three equivalence
classes of binary quadratic forms in C(d). These three classes can be represented by the three forms
Q0 (x, y) = x2 + xy + 6y 2 ,
Q1 (x, y) = 2x2 + xy + 3y 2 ,
2
(4.1)
2
Q2 (x, y) = 2x − xy + 3y .
As shown in Example 2.17, none of these forms are property equivalent. Thus, we can tell that
these three forms are representatives of the three different classes in C(−23). However, we saw in
Example 2.4 that Q1 is equivalent to Q2 , so Q1 and Q2 represent the same integers. Thus, the
theta series of Q1 and Q2 are equal, so we find that our three representative forms have only two
distinct theta series, θQ0 and θQ1 . Using the fact that w−23 = 2 and D−23 = −23, by (3.15) and
(3.17), we have


∞
X
X
1
−23  n
3

fχ0 (z) = (θQ0 (z) + 2θQ1 (z)) = +
q .
2
2
b
n=1
b|n
Suppose χ is one of the two non-trivial characters on Hd . Since hd = 3, χ must take on values
z that satisfy z 3 = 1. First notice that using the isomorphism from Theorem 2.32, we know that
the quadratic form Q0 (x, y) = x2 + xy + 6y 2 gets mapped to the ideal class containing the ideal
√
−1 + −23
[1,
] = O−23 = (1).
2
Thus, Q0 gets mapped to the identity of the ideal class group, so χ(Q0 ) = 1. Also, χ will take
values e±2πi/3 on Q1 and Q2 . Using these results along with (3.10), we see that in this case we have
that fχ = 21 (θQ0 − θQ1 ) which is a Hecke eigenform in the space of cusp forms over the group Γ0 (23)
31
(Kronecker symbol). By Theorem 3.11, if the Fourier expansion of
with characterP−23 (n) = −23
n
∞
n
fχ is fχ (z) = n=0 an q , then its L-series has the form
L(fχ , s) =
Y
p
1 − ap
p−s
1
.
+ −23 (p)p−2s
We also note that for any prime p, −23 (p) equals the Legendre symbol
prime,
p−1
−23
23
= (−1) 2
p
p
p
p−1
p−1 23−1
= (−1) 2 (−1) 2 2
23
p
=
,
23
p 23
because if p is an odd
and by quadratic reciprocity, we additionally have that
2
−23
=1=
.
2
23
Now we can determine ap for each prime p.
Theorem 4.1. Let ap be the Fourier coefficient of q p for the modular form fχ where χ is one of
the nontrivial characters on H−23 . Then

if p =23,

 1

p
= −1,
0
if 23
ap =
(4.2)
p 2
if


23 = 1 and p is represented by Q0 ,

p
−1 if 23 = 1 and p is represented by Q1 .
Proof. Since fχ (z) = 12 (θQ0 − θQ1 ), we have
1
ap = (r(Q0 , p) − r(Q1 , p)).
2
(4.3)
However, we also know
X
r(Q, n) = w
[Q]
X
Dd (b),
b|n
so
p r(Q0 , p) + 2r(Q1 , p) = 2 1 +
.
23
These equations, along with the fact that r(Q0 , p) and r(Q1 , p) must be even (because if (x, y) is a
solution then (−x, −y) is a solution) logically lead us to the conclusion of our theorem.
32
We will next find a connection between the ap from (4.2) and the eta product η(z)η(23z), where
η(z) is as defined in (2.9). We will first prove that the space S1 (Γ0 (23), −23 ) is one-dimensional
and spanned by η(z)η(23z). We will need a few lemmas to reach this conclusion. For brevity, we
will omit the proofs of some lemmas, but we will refer the reader to the source of the lemma.
Q
P
Lemma 4.2. If f (z) = δ|N η(δz)rδ with k = 12 δ|N rδ ∈ Z and with the additional properties
that
X
δrδ ≡ 0 (mod 24)
δ|N
and
XN
δ|N
δ
rδ ≡ 0
(mod 24),
then f (z) ∈ Sk (Γ0 (N ), χ) where χ is defined by χ(d) :=
(−1)k s
d
, where s :=
Q
δ|N
δ rδ .
(See [14], Theorem 1.64.)
Lemma 4.3. Let f (z) = η(z)η(23z) where
η(z)η(23z) = q
∞
Y
(1 − q n )(1 − q 23n ).
(4.4)
n=1
Then f (z) ∈ S1 (Γ0 (23), −23 ).
Proof. We apply Lemma 4.2 with N = 23, and f (z) = η(z)η(23z). In this case, k = 1. We see that
23(1) + 1(1) = 24 ≡ 0
and
23
23
(1) + (1) = 24 ≡ 0
1
23
(mod 24)
(mod 24).
Then we conclude that f (z) ∈ S1 (Γ0 (23), χ) where χ is defined by χ(n) =
−23
n
= −23 (n).
Lemma 4.4. Let k and N be positive integers such that k(N+1)=24. If Sk (Γ1 (N )) is nonzero,
then it is one-dimensional space spanned by η(z)η(23z).
(See [6], Proposition 3.2.2.)
Lemma 4.5. The space Sk (Γ1 (N )) has the following decomposition (where the direct sum is over
all Dirichlet characters modulo N ):
M
Sk (Γ1 (N )) =
Sk (Γ0 (N ), χ).
χ
(See [14], p.5.)
33
Theorem 4.6. The space S1 (Γ0 (23), −23 ) is one-dimensional and is spanned by η(z)η(23z).
Proof. By Lemma 4.3, we know that S1 (Γ0 (23), −23 ) is nonzero. Then since −23 is a Dirichlet
character modulo 23, we know by Lemma 4.5 that S1 (Γ1 (23)) is nonzero. Thus, by Lemma 4.4,
S1 (Γ1 (23)) is one-dimensional and is spanned by η(z)η(23z). Since
M
S1 (Γ1 (23)) =
S1 (Γ0 (23), χ),
χ
and η(z)η(23z) ∈ S1 (Γ0 (23), −23 ), we must also have that S1 (Γ0 (23), −23 ) is one-dimensional and
spanned by η(z)η(23z).
In the next section, it will be proved that the ap from (4.2) are equal to the Fourier coefficients
of q p in the product (4.4).
4.1
A Theorem of van der Blij
Let Q0 , Q1 and Q2 be as defined in (4.1). From (3.16) and (3.17), we have the relation
X −23 r(Q0 , n) + 2r(Q1 , n) = 2
,
d
d|n
noting that 2r(Q1 , n) = r(Q1 , n) + r(Q2 , n) since Q1 and Q2 are equivalent. However, we already
saw that for any prime p,
−23
p
=
.
p
23
Additionally,
−23
1
=1=
1
23
.
Then for each positive integer n, we have that
−23
n
=
.
n
23
Thus,
X d r(Q0 , n) + 2r(Q1 , n) = 2
.
23
(4.5)
d|n
Along with this observation, we will also need the following lemma to prove the theorem of van der
Blij.
34
Lemma 4.7. (The Pentagonal Number Theorem)
For any number x,
∞
Y
(1 − xk ) =
∞
X
1
(−1)m x 2 m(3m+1) .
m=−∞
k=1
We omit the proof of this lemma for brevity. (See [7], p. 225.)
Now we are prepared to prove one of our main theorems, Theorem 1.1. The proof we provide
is as given by van der Blij in [18].
Proof. We will use the identity from Lemma 4.7
∞
Y
(1 − xk ) =
∞
X
1
(−1)m x 2 m(3m+1) .
m=−∞
k=1
With some manipulation, we can see that
∞
Y
∞
X
1
(1 − xk ) = x− 24
1
2
(−1)m x 24 (6m+1) .
m=−∞
k=1
Then using x = q 23 , we have
∞
Y
23
(1 − q 23k ) = q − 24
∞
X
23
2
(−1)m q 24 (6m+1) .
m=−∞
k=1
Now we have that
∞
X
−1
t(n)q n = q(q 24 )(q
−23
24
)
∞
X
1
(−1)m q 24 (6m+1)
m=−∞
n=1
=
∞
X
∞
X
2
23
(−1)p q 24 (6p+1)
2
p=−∞
1
(−1)m+p q 24 (6m+1)
2 + 23 (6p+1)2
24
.
(4.6)
m,p=−∞
Let u = 6m + 1 and v = 6p + 1. Using (4.6), we can see that the term q n appears whenever
u2 + 23v 2 = 24n and m + p is even which occurs when u ≡ v ≡ 1 (mod 12) or u ≡ v ≡ 7 (mod 12).
We can also see that the term −q n appears whenever u2 + 23v 2 = 24n and m + p is odd which
occurs when u ≡ 1, v ≡ 7 or u ≡ 7, v ≡ 1 (mod 12). We will count the number of times q n and
−q n occur, and then we will be able to deduce the number t(n).
First we claim that r(Q0 , n) is the number of solutions (u, v) of u2 + 23v 2 = 24n when u ≡ v
(mod 12). We can see this by noting
24Q0 (x, y) = 24(x2 + xy + 6y 2 )
= (x + 12y)2 + 23x2 .
35
Second we claim that r(Q1 , n) is the number of solutions (u, v) of u2 + 23v 2 = 24n with u ≡ 5v
(mod 12). To prove this, we compare r(Q1 , n) = 2x2 +xy+3y 2 = n and (5x+7y)2 +23(x−y)2 = 24n.
When the latter equation is expanded and simplified, it directly implies the former equation. Also,
we see from the latter equation that u = 5x + 7y and v = x − y. Combining these, we have
u = 5v + 12y which implies that u ≡ 5v (mod 12).
Now, along with the above two claims, we will use the fact that the number of solutions (u, v)
congruent (a, b) and the number of those congruent (−a, b), (a, −b), and (−a, −b) must all be equal.
For example, if u ≡ v ≡ 2 (mod 12), then r(Q1 , n) = r(Q0 , n) since in the second claim we will
have u ≡ 10, v ≡ 2 (mod 12), and (10, 2) ≡ (−2, 2) (mod 12). After all possible combinations are
considered, we can see that
r(Q0 , n) − r(Q1 , n) = 2t(n).
(4.7)
Now combining (4.5) and (4.7), we obtain (1.2) and (1.3).
We are now prepared to prove that the ap from (4.2) are equal to the Fourier coefficients of q p
in the product (4.4).
Corollary 4.8. Let

1



0
ap =
2



−1
if p =23,
p
if 23
= −1,
p if 23 = 1 and p is represented by Q0 ,
p
if 23
= 1 and p is represented by Q1 .
(4.8)
Then ap is equal to the Fourier coefficient of q p for the modular form η(z)η(23z).
P∞
n
p
Proof. Let η(z)η(23z) =
n=1 t(n)q . Then t(p) is the Fourier coefficient of q in the Fourier
expansion of η(z)η(23z). By Theorem 1.1, we have that
2X d
4
r(Q0 , p) =
+ t(p)
3
23
3
d|p
=
2 4
+ t(p),
3 3
(4.9)
and
r(Q1 , p) =
2X
3
d|p
=
d
23
2 2
− t(p).
3 3
2
− t(p)
3
(4.10)
Subtracting (4.9) and (4.10), we have that r(Q0 , p) − r(Q1 , p) = 2t(p). As seen in (4.3), ap =
− r(Q1 , p)). Thus, we can see that t(p) = ap .
1
2 (r(Q0 , p)
36
Chapter 5
Applications and Conclusion
5.1
Ramanujan’s Tau Function
Ramanujan’s Tau Function appears in the Fourier expansion of the cusp form ∆(z) seen in Example
2.44.
Q
24
Recall that for all z ∈ H, ∆(z) = q ∞
n=1 (1 − q) , and ∆(z) is a cusp form of weight 12 on
SL2 (Z). Ramanujan’s tau function τ (n) is defined by
∆(z) =
∞
X
τ (n)q n ,
n=1
where q = e2πiz .
It is easy to see given the definition of ∆(z) that τ (n) is integral for all n. There are also
congruences known about τ (n). For example, it is known that
1 (mod 2) if n is an odd square,
τ (n) ≡
0 (mod 2) otherwise,
and interestingly,
τ (n) ≡ σ11 (n)
where σ11 (n) =
P
11
d|n d
(mod 691) for all n ≥ 1
(see [4], p. 24).
Next we will see that there is a relationship between the ap from (4.2) and τ (p). To see this,
note that for each n,
n n
n 2n
n 24
(1 − q ) = 1 −
q +
q − · · · − 24q 23n + q 24n .
1
2
37
Then, (1 − q n )24 ≡ 1 − q n − q 23n + q 24n (mod 23). Then we have
q
∞
Y
n
(1 − q )(1 − q
23n
)≡q
n=1
∞
Y
(1 − q n )24
(mod 23).
n=1
Since ap is the coefficient of q p in q
Q∞
n=1 (1
− q n )(1 − q 23n ), we have that
τ (p) ≡ ap
(mod 23).
Thus, the modular form η(z)η(23z) reveals something about Ramanujan’s tau function.
5.2
Modular forms and questions of number theory
We√have already seen a connection between the modular form η(z)η(23z) and
√ the quadratic field
Q( −23). However, we can learn even more
prime ideals of Q( −23) by considering
P about the
n . In particular, we will see that the number
the the coefficients ap of q p in η(z)η(23z) = ∞
a
q
n
n=1
ap will indicate whether or not the ideal (p) splits into principal or nonprincipal prime ideals.
We are now prepared to prove our second main theorem, Theorem 1.2.
Proof. Recall that using the isomorphism from Theorem 2.32, Q0 gets mapped to the ideal class
that consists of the principal ideals of O−23 . Let Ai be the ideal class that corresponds to the
quadratic form Qi for each i.
Recall from Section 2.2 that if the norm of an ideal, A, is a prime number, p, then A|(p) as
ideals. Recall also from Section 2.2 that for prime p,

−23

P
if
= −1,



p 0
(p) = P P 0 if −23
p = 1, where P 6= P ,


 2

= 0,
P
if −23
p
where each ideal on the right-hand side is prime, and
ap = r(A0 , p) − r(A1 , p).
Then if ap = 1, then p = 23 and r(A0 ,p) = 1, so we know that (23) splits into the square of a
p
principal prime ideal. If ap = 0, then 23
= −1 and we know that (p) does not split. If ap = 2,
p
then 23 = 1 and r(A0 , p) =2, so (p) splits into the product of two distinct principal prime ideals.
p
Finally, if ap = −1, then 23
= 1 and r(A0 , p) = 0, so (p) splits into the product of two distinct
nonprincipal prime ideals.
38
5.3
Conclusion
Now we have experienced the full collision of binary quadratic forms, quadratic fields, and modular
forms. In Chapter 2, we saw the basics of these three areas and the isomorphism from C(d) to
Hd that connected binary quadratic forms and quadratic fields. Theta series were introduced in
Chapter 3 which provided a connection between binary quadratic forms and modular forms. In
Chapters 4 and 5, we saw these connections all unfold through the specific case d = −23. In
particular, we saw how the representatives of C(−23) as given in (4.1) gave us enough information
to determine the Fourier coefficients of q p , for odd prime p, of the modular form η(z)η(23z). We
then concluded by using these Fourier coefficients to determine information about the ideals of the
ring of integers O−23 .
39
Bibliography
[1] A.N. Andrianov and V.G. Zhuravlev. Modular forms and Hecke operators. Amer Mathematical
Society, 1995.
[2] T.M. Apostol. Introduction to analytic number theory. Springer, 1986.
[3] S. Arno, M.L. Robinson, and F. Wheeler. Imaginary quadratic fields with small odd class
number. Acta Arithmetica, 83:295–330, 1998.
[4] J.H. Bruinier, G. van der Geer, G. Harder, and D. Zagier. The 1-2-3 of modular forms.
Springer, 2008.
[5] D.A. Cox. Primes of the form x2 + ny 2 : Fermat, class field theory, and complex multiplication.
Wiley, 1989.
[6] F. Diamond and J.M. Shurman. A first course in modular forms. Springer Verlag, 2005.
[7] W. Dunham. The genius of Euler: reflections on his life and work. The Mathematical Association of America, 2007.
[8] K.F. Ireland and M.I. Rosen. A classical introduction to modern number theory. Springer,
1990.
[9] Matthew Johnson.
Hecke Eigenforms as Products
http://math.arizona.edu/∼johnsoma/Master’s%20Thesis.pdf.
of
Eigenforms,
2008.
[10] Franz Lemmermeyer. Class Field Theory, 2007. http://www.fen.bilkent.edu.tr/∼franz/cft/cfb.pdf.
[11] S.J. Leon. Linear algebra with applications. Prentice Hall Upper Saddle River, NJ, 2006.
[12] D.A. Marcus. Number fields. Springer, 1977.
[13] H.P. McKean and V. Moll. Elliptic curves: function theory, geometry, arithmetic. Cambridge
Univ Pr, 1999.
[14] K. Ono. The web of modularity: arithmetic of the coefficients of modular forms and q-series.
National Science Foundation, 2004.
[15] A.A. Panchishkin. Introduction to modern number theory: fundamental problems, ideas and
theories. Springer Verlag, 2005.
40
[16] I. Stewart and D.O. Tall. Algebraic number theory and Fermat’s last theorem. AK Peters Ltd,
2002.
[17] A. Straub.
Hermite Normal Form Introduction and Basic Application,
http://arminstraub.com/files/hermitenormalform.pdf.
2007.
[18] F. van der Blij. Binary quadratic forms of discriminant −23. Indagationes Mathematicae,
1952.
[19] T. Weston. Lectures on the Dirichlet Class Number Formula for Imaginary Quadratic Fields.
http://www.math.umass.edu/∼weston/oldpapers/cnf.pdf.
[20] D.B. Zagier. Zetafunktionen und quadratische Körper: eine Einführung in die höhere Zahlentheorie. Springer, 1981.
41